ECAL GPU vs. CPU validation configuration changes #49965
Conversation
…lpaka instead of legacy vs. Alpkaka. Use gpuValidationEcal modifier for the later instead.
|
cms-bot internal usage |
|
+code-checks Logs: https://cmssdt.cern.ch/SDT/code-checks/cms-sw-PR-49965/47746
|
|
A new Pull Request was created by @thomreis for master. It involves the following packages:
@Moanwar, @cmsbuild, @ctarricone, @gabrielmscampos, @jfernan2, @mandrenguyen, @nothingface0, @rseidita, @srimanob can you please review it and eventually sign? Thanks. cms-bot commands are listed here |
|
type ecal |
| Uncalib2DJitterError = _ecalGpuTask.MEs.Uncalib2DJitterError.clone(path = _hltdir + '%(subdet)s/UncalibRecHits/%(prefix)sGT uncalib rec hit jitterError gpu-cpu map2D'), | ||
| Uncalib2DChi2 = _ecalGpuTask.MEs.Uncalib2DChi2.clone(path = _hltdir + '%(subdet)s/UncalibRecHits/%(prefix)sGT uncalib rec hit chi2 gpu-cpu map2D'), | ||
| Uncalib2DOOTAmp = _ecalGpuTask.MEs.Uncalib2DOOTAmp.clone(path = _hltdir + '%(subdet)s/UncalibRecHits/%(prefix)sGT uncalib rec hit OOT amplitude %(OOTAmp)s gpu-cpu map2D'), | ||
| Uncalib2DFlags = _ecalGpuTask.MEs.Uncalib2DFlags.clone(path = _hltdir + '%(subdet)s/UncalibRecHits/%(prefix)sGT uncalib rec hit flags gpu-cpu map2D') |
There was a problem hiding this comment.
what if new plots are added in the cloned PSet? Should this be made automatic?
There was a problem hiding this comment.
If new plots are added they will be going to the default path. But since all variables of the data formats are covered already I do not see why new plots should be added.
But if you have a suggestion to make the changes automatic (add _hltdir and remove %(prefix)sGpuTask/) that would surely be an improvement.
There was a problem hiding this comment.
This seems to work:
_hltdir = 'HLT/HeterogeneousComparison/'
_remove = '%(prefix)sGpuTask/'
def cloneMEsWithPathFix(srcMEs, prefix):
clones = {}
for name, me in srcMEs.parameters_().items():
if hasattr(me, 'path'):
old = me.path.value()
# remove the unwanted component if present
new = old.replace(_remove, '')
# prepend the HLT directory if not already there
new = prefix + new
clones[name] = me.clone(path = new)
else:
clones[name] = me.clone()
return clones
hltEcalGpuTask = _ecalGpuTask.clone(
params = _ecalGpuTask.params.clone(
runGpuTask = True,
enableRecHit = False
),
MEs = cloneMEsWithPathFix(_ecalGpuTask.MEs, _hltdir)
)By the way I noticed that in your earlier implementation the ones that start with RecHit2DChi2 are not touched. Is it done purposefully?
There was a problem hiding this comment.
Great, thanks I'll put that in.
I did not change the RecHit monitoring paths because they are switched off with enableRecHit = False.
|
@cmsbuild please test |
|
-1 Failed Tests: UnitTests RelVals RelVals-INPUT AddOn Failed Unit TestsI found 2 errors in the following unit tests: ---> test test_MC_22_crosscheck had ERRORS ---> test test_MC_23_crosscheck had ERRORS Failed RelVals
Expand to see more relval errors ...
Failed RelVals-INPUT
Expand to see more relval errors ...
Failed AddOn TestsExpand to see more addon errors ... |
cmsbuild
added
tests-pending
code-checks-pending
and removed
tests-rejected
code-checks-approved
labels
|
+code-checks Logs: https://cmssdt.cern.ch/SDT/code-checks/cms-sw-PR-49965/47778 |
|
Pull request #49965 was updated. @Moanwar, @cmsbuild, @ctarricone, @gabrielmscampos, @jfernan2, @mandrenguyen, @nothingface0, @rseidita, @srimanob can you please check and sign again. |
|
enable gpu |
|
+code-checks Logs: https://cmssdt.cern.ch/SDT/code-checks/cms-sw-PR-49965/47779 |
|
Pull request #49965 was updated. @Moanwar, @cmsbuild, @ctarricone, @gabrielmscampos, @jfernan2, @mandrenguyen, @nothingface0, @rseidita, @srimanob can you please check and sign again. |
|
please test |
|
-1 Failed Tests: RelVals-AMD_MI300X Failed RelVals-AMD_MI300X
Comparison SummarySummary:
|
|
The errors look unrelated to the changes in this PR. |
4 participants
PR description:
HLT/HeterogeneousComparison/DQM directory.alpakaValidationEcalmodifier is active. The legacy vs. Alpaka comparison can be activated with thegpuValidationEcalmodifier instead, which also takes care that the timing algorithms match.PR validation:
Passes 18446.413.